Afghanistan: Examining Rare Events

img Taliban attack Farah City

At different points in time in 2018, the Taliban attacked two Afghan cities: Farah City and Ghazni City.1 2 Aside from rare event intrigue, what explains their timing? I use records of violent events3 collected by Afghanistan’s Ministry of Interior to probe for answers. In this notebook, I examine the data, simplify its interpretation through aggregate statistics, and use these statistics to visualize the distribution of the threat in the country.

Findings

Data Exploration

We can see that the dataset is mostly complete in terms of events but is notably incomplete in terms of casualty information.

StatisticNMeanSt. Dev.MinPctl(25)Pctl(75)Max
Tracking.Number10,99238,601.6503,174.12033,13835,844.841,349.244,110
Day10,99215.4228.779182331
Latitude10,99233.9751.51529.43332.74534.87138.409
Longitude10,99267.2662.68461.05864.90069.25471.599
KIA.CIV.NGO.ASG6043.8518.3041.0001.0003.000103.000
ABD.CIV.NGO.ASG499.20425.3751.0001.0005.000170.000
KIA1,9603.0304.1161.0001.0003.00045.000
WIA2,4632.6203.1511.0001.0003.00070.000
ABD.Report.Host.Nation.Security.Military803.7374.2151.0001.0005.00025.000
KIA.Report.Host.Nation.Government351.2290.5471.0001.0001.0003.000
WIA.Report.Host.Nation.Government191.7891.1821.0001.0002.0005.000
ABD.Report.Host.Nation.Government622.16750.3841.0001.0002.750125.000
KIA.Enemy..VEO.Insurgent.Criminal.2,4219.17813.1911.0002.00011.000300.000
WIA.Enemy..VEO.Insurgent.Criminal.1,0716.8707.7791.0002.0009.00082.000
DET.Enemy..VEO.Insurgent.Criminal.1975.2997.1751.0001.0006.00050.000

How Violent is the Country?

The barplot below shows some clustering in values, suggesting that we can further classify our provinces into easy to understand threat categories like: low, moderate, and high. Percentiles is a good solution.



With the below categories specified, the barplot is presented again in slightly different fashion. We can see that the use of percentiles does a fairly good job of classifying provinces into simplified threat categories.


\[\text{Province Threat} = \begin{cases} \text{high threat:} \hspace{1.5cm} \text{violent events} > 75\% \\ \text{moderate threat:} \hspace{.7cm} 25\% > \text{violent events} < 75\% \\ \text{low threat:} \hspace{1.7cm} \text{violent events} < 25\% \\ \end{cases} \]



A Geospatial Look

While the barplots are interesting, they offer little insight into whether or not there is a geospatial component to the data. The map below fixes that. We can see that there is a substantive difference in the levels of violent of events depending on where you live in the country, with the south having the most violent events.



  1. https://www.nytimes.com/2018/05/16/world/asia/taliban-farah-afghanistan.html

  2. https://time.com/longform/ghazni-fight-taliban/

  3. This includes government actions, insurgent actions, and explosions.

  4. High threat provinces were determined by separating the violence levels of provinces into percentiles, with high threat provinces falling into the 75%. This means that 75% of all provinces have fewer violent events than high threat provinces.